microsoft/DeepSpeed

Pull Requests Commits

Use accelerator pin memory

tjruwase committed 77 days ago

970955dc

fix: engine initializes optimizer attributes at the beginning (#7410)

HollowMan6 committed 78 days ago

Verified 15f054d9

fix: fix FileNotFoundError for build_win.bat (#7399)

gjj2828 committed 83 days ago

Verified da60a878

fix(comm): Expose GradBucket in deepspeed.comm API (#7400)

Flink-ddd committed 84 days ago

Verified e6324af9

fix: DeepCompile for torch 2.8 (#7402)

HollowMan6 committed 84 days ago

Verified 384e6d34

fix: Add `csrc/compile` to include paths for DeepCompile builder (#7401)

HollowMan6 committed 84 days ago

Verified a755a9e5

Added device detection to communication logging (#7398)

alexk101 committed 86 days ago

Verified 4c687bfd

Improve coverage of DeepCompile (#7386)

tohtana committed 87 days ago

Verified 6594c266

Fix ZeRO stage 1 and add stage 2 support with DeepCompile (#7366)

tohtana committed 87 days ago

Verified be8124c8

Fix unbound local error for `return_val` (#7395)

HollowMan6 committed 87 days ago

Verified 6c469425

add support for CUDAtk12.9 (#7394)

loscrossos committed 87 days ago

Verified 59bb08bf

fix broken url (#7390)

stas00 committed 88 days ago

Verified ebb64239

add blog link (#7385)

stas00 committed 90 days ago

Verified ec73d91b

fix(inference): Add missing dtype attribute to ParameterBase setter (#7378)

Flink-ddd committed 91 days ago

Verified 61829b55

Add support for ws=1 scenario (#7379)

NirSonnenschein committed 91 days ago

Verified 2a450b3a

Fix dtype mismatch in `TestParamPartitioningSkipInit` (#7377)

tohtana committed 91 days ago

Verified e049bbfa

fix wandb.log() call by removing `sync` kwarg (#7383)

ned2 committed 91 days ago

Verified 8dd21516

Fix release of IPG buffer (#7376)

tohtana committed 93 days ago

Verified d5f69151

Update latest news with DeepNVMe (#7375)

loadams committed 94 days ago

Verified 9606f8f0

Relax tolerances for FP8 unit test only for ROCm + FP16 (#7373)

rraminen committed 94 days ago

Verified d33baf00

Flops profiler support for F.interpolate (#7353)

sfc-gh-truwase committed 94 days ago

Verified 25da6fc1

add ALST paper reference (#7372)

stas00 committed 95 days ago

Verified 86097872

Enable torch.autocast with ZeRO (#6993)

tohtana committed 95 days ago

Verified ed5f7375

sequence parallel default dtype (#7364)

stas00 committed 95 days ago

Verified d3b9cb8c

Fix(scheduler): WarmupLR inherits optimizer lr when not specified (#7360)

Flink-ddd committed 95 days ago

Verified 22cf1a44

Restore real inputs for recompilation (#7356)

tohtana committed 95 days ago

Verified 6f1a1c04

Fix tutorial title (#7365)

stas00 committed 97 days ago

Verified f394e780

Fix 404s (#7363)

tjruwase committed 98 days ago

Verified 9ac94414

Improve padding util for compile (#7355)

tohtana committed 100 days ago

Verified 600d280f

Fix error of <glog/logging.h> (#7351)

Freed-Wu committed 102 days ago

Verified 76631215

Older